Apparatus and method for performing artificial intelligence encoding and artificial intelligence decoding of image

Abstract

A method, performed by a terminal, of performing artificial intelligence (AI) decoding, including obtaining, based on a first image, image feature information of the first image, the image feature information being related to an image quality degradation; obtaining, based on the image feature information of the first image, neural network (NN) setting information of a first NN, from among NN setting information of a plurality of first NNs which are pre-stored and which correspond to a plurality of image quality degradation types; and obtaining, by using the first NN, a second image in which the image quality degradation is reduced.

The invention claimed is: 
     
       1. A method, performed by a terminal, of performing artificial intelligence (AI) decoding, the method comprising:
 obtaining, based on a first image, image feature information of the first image, the image feature information being related to an image quality degradation; 
 obtaining, based on the image feature information of the first image, neural network (NN) setting information of a first NN corresponding to an image quality degradation type of the first image, from among NN setting information of a plurality of first NNs which are pre-stored, wherein each first NN from among the plurality of first NNs corresponds to an image quality degradation type from among a plurality of image quality degradation types; and 
 obtaining, by using the first NN, a second image in which the image quality degradation is reduced. 
 
     
     
       2. The method of  claim 1 , wherein the image feature information is obtained by inputting the first image to a second NN. 
     
     
       3. The method of  claim 2 , wherein a first image data set is obtained through image crawling,
 wherein a second image data set comprising a plurality of training images, which include a same object and have different image quality degradation elements, is obtained from the first image data set, 
 wherein one training image of the plurality of training images included in the second image data set is identified as a label image, at least one training image corresponding to a preset image quality degradation type is identified among the plurality of training images other than the label image, 
 wherein the label image and the at least one training image corresponding to the preset image quality degradation type are identified as a training image data set, and 
 wherein the second NN for identifying the preset image quality degradation type is trained by using the training image data set. 
 
     
     
       4. The method of  claim 2 , wherein the image quality degradation comprises at least one of resolution-related image quality degradations, color-related image quality degradations, brightness-related image quality degradations, or encoding-related image quality degradations. 
     
     
       5. The method of  claim 1 , wherein the obtaining of the NN setting information of the first NN comprises:
 transmitting the image feature information of the first image to a server; and 
 receiving, from the server, based on the image feature information of the first image, the NN setting information of the first NN from among the NN setting information of the plurality of first NNs which are pre-stored in the server. 
 
     
     
       6. The method of  claim 1 , wherein a first image data set is obtained through image crawling,
 wherein a second image data set comprising a plurality of training images, which include a same object and which have different image quality degradation elements, is obtained from the first image data set, 
 wherein one training image of the plurality of training images included in the second image data set is identified as a label image, at least one training image corresponding to a preset image quality degradation type is identified among the plurality of training images other than the label image, and the label image and the at least one training image corresponding to the preset image quality degradation type are included in a training image data set, and 
 wherein one first NN from among the plurality of first NNs for processing an image corresponding to the preset image quality degradation type is trained by using the training image data set. 
 
     
     
       7. The method of  claim 6 , wherein, when the second image data set is obtained from the first image data set,
 wherein feature points of images included in the first image data set are identified, and 
 wherein the plurality of training images are identified based on the identified feature points. 
 
     
     
       8. The method of  claim 1 , wherein the obtaining of the NN setting information of the first NN comprises:
 obtaining, based on the image feature information of the first image, a value indicating a degree of a preset image quality degradation type present in the first image; 
 based on the value being greater than a first threshold, determining to perform image processing corresponding to the preset image quality degradation type by using the first NN; and 
 based on the value indicating the degree of the preset image quality degradation type being less than the first threshold, determining not to perform the image processing corresponding to the preset image quality degradation type by using the first NN. 
 
     
     
       9. The method of  claim 1 , wherein the obtaining of the image feature information of the first image, the obtaining of the NN setting information of the first NN, and the obtaining of the second image in which the image quality degradation is reduced are repeatedly performed with regard to the plurality of image quality degradation types. 
     
     
       10. The method of  claim 1 , wherein the obtaining of the NN setting information of the first NN comprises:
 transmitting the image feature information of the first image to a server; 
 receiving, from the server, flag information indicating whether to perform AI image processing, the flag information being determined based on the image feature information of the first image; and 
 obtaining the NN setting information of the first NN based on the flag information. 
 
     
     
       11. The method of  claim 1 , wherein the obtaining of the NN setting information of the first NN comprises:
 obtaining, based on the image feature information of the first image, values indicating degrees of the plurality of image quality degradation types of the first image; 
 identifying an image quality degradation type having a highest value among the values; and 
 obtaining the NN setting information of the first NN corresponding to the image quality degradation type among the NN setting information of the plurality of first NNs. 
 
     
     
       12. The method of  claim 1 , wherein the image feature information is obtained based on at least one of a color histogram of the first image, edges and blocking artifacts detected in the first image, a bit rate of the first image, or a resolution of the first image. 
     
     
       13. The method of  claim 1 , further comprising:
 selecting, based on the image feature information of the first image, the image quality degradation type of the first image from the plurality of image quality degradation types. 
 
     
     
       14. A terminal for artificial intelligence (AI) decoding, the terminal comprising at least one processor configured to execute one or more instructions to:
 obtain, based on a first image, image feature information of the first image, the image feature information being related to an image quality degradation, 
 obtain, based on the image feature information of the first image, neural network (NN) setting information of a first NN corresponding to an image quality degradation type of the first image, from among NN setting information of a plurality of first NNs which are pre-stored, wherein each first NN from among the plurality of first NNs corresponds to an image quality degradation type from among a plurality of image quality degradation types, and 
 obtain, by using the first NN, which is set based on the NN setting information of the first NN, a second image in which the image quality degradation is reduced. 
 
     
     
       15. The terminal of  claim 14 , wherein the image feature information is obtained by inputting the first image to a second neural network. 
     
     
       16. A non-transitory computer-readable recording medium having recorded thereon instructions which, when executed by at least one processor, cause the at least one processor to:
 obtain, based on a first image, image feature information of the first image, the image feature information being related to an image quality degradation; 
 obtain, based on the image feature information of the first image, neural network (NN) setting information of a first NN corresponding to an image quality degradation type of the first image, from among NN setting information of a plurality of first NNs which are pre-stored, wherein each first NN from among the plurality of first NNs corresponds to an image quality degradation type from among a plurality of image quality degradation types; and 
 obtain, by using the first NN, a second image in which the image quality degradation is reduced.